Implementation of a Deep Learning Technique for Detection of Objects on Roads for Autonomous Vehicles

Autonomous driving and vehicle safety things profoundly depend on object detection to guarantee precise an robust right-time performance under different environmental conditions. The effectiveness of this systems depends on precise and efficient processing algorithms. The manuscript aim is to contrast the correctness of three notable image processing algorithms in the context of auto driving: Convolutional Neural Networks (CNN), Retina Net, and You Only Look Once v8 (YOLOv8). The assessment is performed using the Roboflow and kaggle dataset. The first algorithm under assessment is the Convolutional Neural Networks(CNN), Retina Net, and You Only Look Once v8 (YOLOv8). Each algorithm's unique approach to object detection is reviewed, observing their strengths and weaknesses in accurately identifying things in complex road situations. A comparative analysis is performed using the Robo flow and Kaggle dataset, covering various real-world contexts. The algorithms are evaluated based on correctness, considering occurrences with and without occlusion and truncation. Computation time,a vital aspect for real-time applications, is also considered. The precision-recall curve is applied as a metric to determine the algorithms' overall performance. Retina Net, with its focal loss mechanism, is reviewed for its capability to manage occluded and truncated things and its influence on overall correctness. YOLOv8 emerges as a standout performer, displaying superior correctness in discovering challenging road target things under complex situations and adverse weather conditions. Its right-time deployment capabilities, combined with high correctness, position it as a compelling choice for autonomous driving systems. In conclusion, this comparative analysis shines a light on the strengths and deficiencies of notable image processing algorithms for object discovery in autonomous driving systems. YOLOv8 stands outas the most accurate and efficient algorithm, excelling in challenging real-world situations. As the demand for reliable and robust autonomous driving systems grows, understanding the nuances of these algorithms becomes paramount for informed decision-making in real-time deployment restrictions